/*

 * Copyright (c) 2012, 2019, Oracle and/or its affiliates. All rights reserved.

 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.

 *

 * This code is free software; you can redistribute it and/or modify it

 * under the terms of the GNU General Public License version 2 only, as

 * published by the Free Software Foundation.

 *

 * This code is distributed in the hope that it will be useful, but WITHOUT

 * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or

 * FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public License

 * version 2 for more details (a copy is included in the LICENSE file that

 * accompanied this code).

 *

 * You should have received a copy of the GNU General Public License version

 * 2 along with this work; if not, write to the Free Software Foundation,

 * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA.

 *

 * Please contact Oracle, 500 Oracle Parkway, Redwood Shores, CA 94065 USA

 * or visit www.oracle.com if you need additional information or have any

 * questions.

 *

 */

#include "precompiled.hpp"

#include "gc/shared/barrierSet.hpp"

#include "opto/arraycopynode.hpp"

#include "oops/objArrayKlass.hpp"

#include "opto/convertnode.hpp"

#include "opto/graphKit.hpp"

#include "opto/macro.hpp"

#include "opto/runtime.hpp"

#include "utilities/align.hpp"

void PhaseMacroExpand::insert_mem_bar(Node** ctrl, Node** mem, int opcode, Node* precedent) {

  MemBarNode* mb = MemBarNode::make(C, opcode, Compile::AliasIdxBot, precedent);

  mb->init_req(TypeFunc::Control, *ctrl);

  mb->init_req(TypeFunc::Memory, *mem);

  transform_later(mb);

  *ctrl = new ProjNode(mb,TypeFunc::Control);

  transform_later(*ctrl);

  Node* mem_proj = new ProjNode(mb,TypeFunc::Memory);

  transform_later(mem_proj);

  *mem = mem_proj;

}

Node* PhaseMacroExpand::array_element_address(Node* ary, Node* idx, BasicType elembt) {

  uint shift  = exact_log2(type2aelembytes(elembt));

  uint header = arrayOopDesc::base_offset_in_bytes(elembt);

  Node* base =  basic_plus_adr(ary, header);

#ifdef _LP64

  // see comment in GraphKit::array_element_address

  int index_max = max_jint - 1;  // array size is max_jint, index is one less

  const TypeLong* lidxtype = TypeLong::make(CONST64(0), index_max, Type::WidenMax);

  idx = transform_later( new ConvI2LNode(idx, lidxtype) );

#endif

  Node* scale = new LShiftXNode(idx, intcon(shift));

  transform_later(scale);

  return basic_plus_adr(ary, base, scale);

}

Node* PhaseMacroExpand::ConvI2L(Node* offset) {

  return transform_later(new ConvI2LNode(offset));

}

Node* PhaseMacroExpand::make_leaf_call(Node* ctrl, Node* mem,

                                       const TypeFunc* call_type, address call_addr,

                                       const char* call_name,

                                       const TypePtr* adr_type,

                                       Node* parm0, Node* parm1,

                                       Node* parm2, Node* parm3,

                                       Node* parm4, Node* parm5,

                                       Node* parm6, Node* parm7) {

  Node* call = new CallLeafNoFPNode(call_type, call_addr, call_name, adr_type);

  call->init_req(TypeFunc::Control, ctrl);

  call->init_req(TypeFunc::I_O    , top());

  call->init_req(TypeFunc::Memory , mem);

  call->init_req(TypeFunc::ReturnAdr, top());

  call->init_req(TypeFunc::FramePtr, top());

  // Hook each parm in order.  Stop looking at the first NULL.

  if (parm0 != NULL) { call->init_req(TypeFunc::Parms+0, parm0);

  if (parm1 != NULL) { call->init_req(TypeFunc::Parms+1, parm1);

  if (parm2 != NULL) { call->init_req(TypeFunc::Parms+2, parm2);

  if (parm3 != NULL) { call->init_req(TypeFunc::Parms+3, parm3);

  if (parm4 != NULL) { call->init_req(TypeFunc::Parms+4, parm4);

  if (parm5 != NULL) { call->init_req(TypeFunc::Parms+5, parm5);

  if (parm6 != NULL) { call->init_req(TypeFunc::Parms+6, parm6);

  if (parm7 != NULL) { call->init_req(TypeFunc::Parms+7, parm7);

    /* close each nested if ===> */  } } } } } } } }

  assert(call->in(call->req()-1) != NULL, "must initialize all parms");

  return call;

}

//------------------------------generate_guard---------------------------

// Helper function for generating guarded fast-slow graph structures.

// The given 'test', if true, guards a slow path.  If the test fails

// then a fast path can be taken.  (We generally hope it fails.)

// In all cases, GraphKit::control() is updated to the fast path.

// The returned value represents the control for the slow path.

// The return value is never 'top'; it is either a valid control

// or NULL if it is obvious that the slow path can never be taken.

// Also, if region and the slow control are not NULL, the slow edge

// is appended to the region.

Node* PhaseMacroExpand::generate_guard(Node** ctrl, Node* test, RegionNode* region, float true_prob) {

  if ((*ctrl)->is_top()) {

    // Already short circuited.

    return NULL;

  }

  // Build an if node and its projections.

  // If test is true we take the slow path, which we assume is uncommon.

  if (_igvn.type(test) == TypeInt::ZERO) {

    // The slow branch is never taken.  No need to build this guard.

    return NULL;

  }

  IfNode* iff = new IfNode(*ctrl, test, true_prob, COUNT_UNKNOWN);

  transform_later(iff);

  Node* if_slow = new IfTrueNode(iff);

  transform_later(if_slow);

  if (region != NULL) {

    region->add_req(if_slow);

  }

  Node* if_fast = new IfFalseNode(iff);

  transform_later(if_fast);

  *ctrl = if_fast;

  return if_slow;

}

inline Node* PhaseMacroExpand::generate_slow_guard(Node** ctrl, Node* test, RegionNode* region) {

  return generate_guard(ctrl, test, region, PROB_UNLIKELY_MAG(3));

}

void PhaseMacroExpand::generate_negative_guard(Node** ctrl, Node* index, RegionNode* region) {

  if ((*ctrl)->is_top())

    return;                // already stopped

  if (_igvn.type(index)->higher_equal(TypeInt::POS)) // [0,maxint]

    return;                // index is already adequately typed

  Node* cmp_lt = new CmpINode(index, intcon(0));

  transform_later(cmp_lt);

  Node* bol_lt = new BoolNode(cmp_lt, BoolTest::lt);

  transform_later(bol_lt);

  generate_guard(ctrl, bol_lt, region, PROB_MIN);

}

void PhaseMacroExpand::generate_limit_guard(Node** ctrl, Node* offset, Node* subseq_length, Node* array_length, RegionNode* region) {

  if ((*ctrl)->is_top())

    return;                // already stopped

  bool zero_offset = _igvn.type(offset) == TypeInt::ZERO;

  if (zero_offset && subseq_length->eqv_uncast(array_length))

    return;                // common case of whole-array copy

  Node* last = subseq_length;

  if (!zero_offset) {            // last += offset

    last = new AddINode(last, offset);

    transform_later(last);

  }

  Node* cmp_lt = new CmpUNode(array_length, last);

  transform_later(cmp_lt);

  Node* bol_lt = new BoolNode(cmp_lt, BoolTest::lt);

  transform_later(bol_lt);

  generate_guard(ctrl, bol_lt, region, PROB_MIN);

}

Node* PhaseMacroExpand::generate_nonpositive_guard(Node** ctrl, Node* index, bool never_negative) {

  if ((*ctrl)->is_top())  return NULL;

  if (_igvn.type(index)->higher_equal(TypeInt::POS1)) // [1,maxint]

    return NULL;                // index is already adequately typed

  Node* cmp_le = new CmpINode(index, intcon(0));

  transform_later(cmp_le);

  BoolTest::mask le_or_eq = (never_negative ? BoolTest::eq : BoolTest::le);

  Node* bol_le = new BoolNode(cmp_le, le_or_eq);

  transform_later(bol_le);

  Node* is_notp = generate_guard(ctrl, bol_le, NULL, PROB_MIN);

  return is_notp;

}

void PhaseMacroExpand::finish_arraycopy_call(Node* call, Node** ctrl, MergeMemNode** mem, const TypePtr* adr_type) {

  transform_later(call);

  *ctrl = new ProjNode(call,TypeFunc::Control);

  transform_later(*ctrl);

  Node* newmem = new ProjNode(call, TypeFunc::Memory);

  transform_later(newmem);

  uint alias_idx = C->get_alias_index(adr_type);

  if (alias_idx != Compile::AliasIdxBot) {

    *mem = MergeMemNode::make(*mem);

    (*mem)->set_memory_at(alias_idx, newmem);

  } else {

    *mem = MergeMemNode::make(newmem);

  }

  transform_later(*mem);

}

address PhaseMacroExpand::basictype2arraycopy(BasicType t,

                                              Node* src_offset,

                                              Node* dest_offset,

                                              bool disjoint_bases,

                                              const char* &name,

                                              bool dest_uninitialized) {

  const TypeInt* src_offset_inttype  = _igvn.find_int_type(src_offset);;

  const TypeInt* dest_offset_inttype = _igvn.find_int_type(dest_offset);;

  bool aligned = false;

  bool disjoint = disjoint_bases;

  // if the offsets are the same, we can treat the memory regions as

  // disjoint, because either the memory regions are in different arrays,

  // or they are identical (which we can treat as disjoint.)  We can also

  // treat a copy with a destination index  less that the source index

  // as disjoint since a low->high copy will work correctly in this case.

  if (src_offset_inttype != NULL && src_offset_inttype->is_con() &&

      dest_offset_inttype != NULL && dest_offset_inttype->is_con()) {

    // both indices are constants

    int s_offs = src_offset_inttype->get_con();

    int d_offs = dest_offset_inttype->get_con();

    int element_size = type2aelembytes(t);

    aligned = ((arrayOopDesc::base_offset_in_bytes(t) + s_offs * element_size) % HeapWordSize == 0) &&

              ((arrayOopDesc::base_offset_in_bytes(t) + d_offs * element_size) % HeapWordSize == 0);

    if (s_offs >= d_offs)  disjoint = true;

  } else if (src_offset == dest_offset && src_offset != NULL) {

    // This can occur if the offsets are identical non-constants.

    disjoint = true;

  }

  return StubRoutines::select_arraycopy_function(t, aligned, disjoint, name, dest_uninitialized);

}

#define XTOP LP64_ONLY(COMMA top())

// Generate an optimized call to arraycopy.

// Caller must guard against non-arrays.

// Caller must determine a common array basic-type for both arrays.

// Caller must validate offsets against array bounds.

// The slow_region has already collected guard failure paths

// (such as out of bounds length or non-conformable array types).

// The generated code has this shape, in general:

//

//     if (length == 0)  return   // via zero_path

//     slowval = -1

//     if (types unknown) {

//       slowval = call generic copy loop

//       if (slowval == 0)  return  // via checked_path

//     } else if (indexes in bounds) {

//       if ((is object array) && !(array type check)) {

//         slowval = call checked copy loop

//         if (slowval == 0)  return  // via checked_path

//       } else {

//         call bulk copy loop

//         return  // via fast_path

//       }

//     }

//     // adjust params for remaining work:

//     if (slowval != -1) {

//       n = -1^slowval; src_offset += n; dest_offset += n; length -= n

//     }

//   slow_region:

//     call slow arraycopy(src, src_offset, dest, dest_offset, length)

//     return  // via slow_call_path

//

// This routine is used from several intrinsics:  System.arraycopy,

// Object.clone (the array subcase), and Arrays.copyOf[Range].

//

Node* PhaseMacroExpand::generate_arraycopy(ArrayCopyNode *ac, AllocateArrayNode* alloc,

                                           Node** ctrl, MergeMemNode* mem, Node** io,

                                           const TypePtr* adr_type,

                                           BasicType basic_elem_type,

                                           Node* src,  Node* src_offset,

                                           Node* dest, Node* dest_offset,

                                           Node* copy_length,

                                           bool disjoint_bases,

                                           bool length_never_negative,

                                           RegionNode* slow_region) {

  if (slow_region == NULL) {

    slow_region = new RegionNode(1);

    transform_later(slow_region);

  }

  Node* original_dest      = dest;

  bool  dest_uninitialized = false;

  // See if this is the initialization of a newly-allocated array.

  // If so, we will take responsibility here for initializing it to zero.

  // (Note:  Because tightly_coupled_allocation performs checks on the

  // out-edges of the dest, we need to avoid making derived pointers

  // from it until we have checked its uses.)

  if (ReduceBulkZeroing

      && !(UseTLAB && ZeroTLAB) // pointless if already zeroed

      && basic_elem_type != T_CONFLICT // avoid corner case

      && !src->eqv_uncast(dest)

      && alloc != NULL

      && _igvn.find_int_con(alloc->in(AllocateNode::ALength), 1) > 0

      && alloc->maybe_set_complete(&_igvn)) {

    // "You break it, you buy it."

    InitializeNode* init = alloc->initialization();

    assert(init->is_complete(), "we just did this");

    init->set_complete_with_arraycopy();

    assert(dest->is_CheckCastPP(), "sanity");

    assert(dest->in(0)->in(0) == init, "dest pinned");

    adr_type = TypeRawPtr::BOTTOM;  // all initializations are into raw memory

    // From this point on, every exit path is responsible for

    // initializing any non-copied parts of the object to zero.

    // Also, if this flag is set we make sure that arraycopy interacts properly

    // with G1, eliding pre-barriers. See CR 6627983.

    dest_uninitialized = true;

  } else {

    // No zeroing elimination here.

    alloc             = NULL;

    //original_dest   = dest;

    //dest_uninitialized = false;

  }

  uint alias_idx = C->get_alias_index(adr_type);

  // Results are placed here:

  enum { fast_path        = 1,  // normal void-returning assembly stub

         checked_path     = 2,  // special assembly stub with cleanup

         slow_call_path   = 3,  // something went wrong; call the VM

         zero_path        = 4,  // bypass when length of copy is zero

         bcopy_path       = 5,  // copy primitive array by 64-bit blocks

         PATH_LIMIT       = 6

  };

  RegionNode* result_region = new RegionNode(PATH_LIMIT);

  PhiNode*    result_i_o    = new PhiNode(result_region, Type::ABIO);

  PhiNode*    result_memory = new PhiNode(result_region, Type::MEMORY, adr_type);

  assert(adr_type != TypePtr::BOTTOM, "must be RawMem or a T[] slice");

  transform_later(result_region);

  transform_later(result_i_o);

  transform_later(result_memory);

  // The slow_control path:

  Node* slow_control;

  Node* slow_i_o = *io;

  Node* slow_mem = mem->memory_at(alias_idx);

  DEBUG_ONLY(slow_control = (Node*) badAddress);

  // Checked control path:

  Node* checked_control = top();

  Node* checked_mem     = NULL;

  Node* checked_i_o     = NULL;

  Node* checked_value   = NULL;

  if (basic_elem_type == T_CONFLICT) {

    assert(!dest_uninitialized, "");

    Node* cv = generate_generic_arraycopy(ctrl, &mem,

                                          adr_type,

                                          src, src_offset, dest, dest_offset,

                                          copy_length, dest_uninitialized);

    if (cv == NULL)  cv = intcon(-1);  // failure (no stub available)

    checked_control = *ctrl;

    checked_i_o     = *io;

    checked_mem     = mem->memory_at(alias_idx);

    checked_value   = cv;

    *ctrl = top();

  }

  Node* not_pos = generate_nonpositive_guard(ctrl, copy_length, length_never_negative);

  if (not_pos != NULL) {

    Node* local_ctrl = not_pos, *local_io = *io;

    MergeMemNode* local_mem = MergeMemNode::make(mem);

    transform_later(local_mem);

    // (6) length must not be negative.

    if (!length_never_negative) {

      generate_negative_guard(&local_ctrl, copy_length, slow_region);

    }

    // copy_length is 0.

    if (dest_uninitialized) {

      assert(!local_ctrl->is_top(), "no ctrl?");

      Node* dest_length = alloc->in(AllocateNode::ALength);

      if (copy_length->eqv_uncast(dest_length)

          || _igvn.find_int_con(dest_length, 1) <= 0) {

        // There is no zeroing to do. No need for a secondary raw memory barrier.

      } else {

        // Clear the whole thing since there are no source elements to copy.

        generate_clear_array(local_ctrl, local_mem,

                             adr_type, dest, basic_elem_type,

                             intcon(0), NULL,

                             alloc->in(AllocateNode::AllocSize));

        // Use a secondary InitializeNode as raw memory barrier.

        // Currently it is needed only on this path since other

        // paths have stub or runtime calls as raw memory barriers.

        MemBarNode* mb = MemBarNode::make(C, Op_Initialize,

                                          Compile::AliasIdxRaw,

                                          top());

        transform_later(mb);

        mb->set_req(TypeFunc::Control,local_ctrl);

        mb->set_req(TypeFunc::Memory, local_mem->memory_at(Compile::AliasIdxRaw));

        local_ctrl = transform_later(new ProjNode(mb, TypeFunc::Control));

        local_mem->set_memory_at(Compile::AliasIdxRaw, transform_later(new ProjNode(mb, TypeFunc::Memory)));

        InitializeNode* init = mb->as_Initialize();

        init->set_complete(&_igvn);  // (there is no corresponding AllocateNode)

      }

    }

    // Present the results of the fast call.

    result_region->init_req(zero_path, local_ctrl);

    result_i_o   ->init_req(zero_path, local_io);

    result_memory->init_req(zero_path, local_mem->memory_at(alias_idx));

  }

  if (!(*ctrl)->is_top() && dest_uninitialized) {

    // We have to initialize the *uncopied* part of the array to zero.

    // The copy destination is the slice dest[off..off+len].  The other slices

    // are dest_head = dest[0..off] and dest_tail = dest[off+len..dest.length].

    Node* dest_size   = alloc->in(AllocateNode::AllocSize);

    Node* dest_length = alloc->in(AllocateNode::ALength);

    Node* dest_tail   = transform_later( new AddINode(dest_offset, copy_length));

    // If there is a head section that needs zeroing, do it now.

    if (_igvn.find_int_con(dest_offset, -1) != 0) {

      generate_clear_array(*ctrl, mem,

                           adr_type, dest, basic_elem_type,

                           intcon(0), dest_offset,

                           NULL);

    }

    // Next, perform a dynamic check on the tail length.

    // It is often zero, and we can win big if we prove this.

    // There are two wins:  Avoid generating the ClearArray

    // with its attendant messy index arithmetic, and upgrade

    // the copy to a more hardware-friendly word size of 64 bits.

    Node* tail_ctl = NULL;

    if (!(*ctrl)->is_top() && !dest_tail->eqv_uncast(dest_length)) {

      Node* cmp_lt   = transform_later( new CmpINode(dest_tail, dest_length) );

      Node* bol_lt   = transform_later( new BoolNode(cmp_lt, BoolTest::lt) );

      tail_ctl = generate_slow_guard(ctrl, bol_lt, NULL);

      assert(tail_ctl != NULL || !(*ctrl)->is_top(), "must be an outcome");

    }

    // At this point, let's assume there is no tail.

    if (!(*ctrl)->is_top() && alloc != NULL && basic_elem_type != T_OBJECT) {

      // There is no tail.  Try an upgrade to a 64-bit copy.

      bool didit = false;

      {

        Node* local_ctrl = *ctrl, *local_io = *io;

        MergeMemNode* local_mem = MergeMemNode::make(mem);

        transform_later(local_mem);

        didit = generate_block_arraycopy(&local_ctrl, &local_mem, local_io,

                                         adr_type, basic_elem_type, alloc,

                                         src, src_offset, dest, dest_offset,

                                         dest_size, dest_uninitialized);

        if (didit) {

          // Present the results of the block-copying fast call.

          result_region->init_req(bcopy_path, local_ctrl);

          result_i_o   ->init_req(bcopy_path, local_io);

          result_memory->init_req(bcopy_path, local_mem->memory_at(alias_idx));

        }

      }

      if (didit) {

        *ctrl = top();     // no regular fast path

      }

    }

    // Clear the tail, if any.

    if (tail_ctl != NULL) {

      Node* notail_ctl = (*ctrl)->is_top() ? NULL : *ctrl;

      *ctrl = tail_ctl;

      if (notail_ctl == NULL) {

        generate_clear_array(*ctrl, mem,

                             adr_type, dest, basic_elem_type,

                             dest_tail, NULL,

                             dest_size);

      } else {

        // Make a local merge.

        Node* done_ctl = transform_later(new RegionNode(3));

        Node* done_mem = transform_later(new PhiNode(done_ctl, Type::MEMORY, adr_type));

        done_ctl->init_req(1, notail_ctl);

        done_mem->init_req(1, mem->memory_at(alias_idx));

        generate_clear_array(*ctrl, mem,

                             adr_type, dest, basic_elem_type,

                             dest_tail, NULL,

                             dest_size);

        done_ctl->init_req(2, *ctrl);

        done_mem->init_req(2, mem->memory_at(alias_idx));

        *ctrl = done_ctl;

        mem->set_memory_at(alias_idx, done_mem);

      }

    }

  }

  BasicType copy_type = basic_elem_type;

  assert(basic_elem_type != T_ARRAY, "caller must fix this");

  if (!(*ctrl)->is_top() && copy_type == T_OBJECT) {

    // If src and dest have compatible element types, we can copy bits.

    // Types S[] and D[] are compatible if D is a supertype of S.

    //

    // If they are not, we will use checked_oop_disjoint_arraycopy,

    // which performs a fast optimistic per-oop check, and backs off

    // further to JVM_ArrayCopy on the first per-oop check that fails.

    // (Actually, we don't move raw bits only; the GC requires card marks.)

    // We don't need a subtype check for validated copies and Object[].clone()

    bool skip_subtype_check = ac->is_arraycopy_validated() || ac->is_copyof_validated() ||

    if (!skip_subtype_check) {

      // Get the klass* for both src and dest

      Node* src_klass  = ac->in(ArrayCopyNode::SrcKlass);

      Node* dest_klass = ac->in(ArrayCopyNode::DestKlass);

      assert(src_klass != NULL && dest_klass != NULL, "should have klasses");

      // Generate the subtype check.

      // This might fold up statically, or then again it might not.

      //

      // Non-static example:  Copying List<String>.elements to a new String[].

      // The backing store for a List<String> is always an Object[],

      // but its elements are always type String, if the generic types

      // are correct at the source level.

      //

      // Test S[] against D[], not S against D, because (probably)

      // the secondary supertype cache is less busy for S[] than S.

      // This usually only matters when D is an interface.